Mystery group claims it has been using Claude Mythos since early April

Anthropic is investigating a significant security breach involving Claude Mythos, a model the company previously labeled as too dangerous for public release. An anonymous group reportedly gained unauthorized access to the model by exploiting a data leak from the AI startup Mercor combined with credentials from a third-party contractor. While the group claims they are only interested in "vibe coding" experiments rather than causing harm, the incident raises critical concerns about the safety and internal governance of high-risk frontier AI models.

Read the full story on Gizmodo →